Verified, defensible investigation of complex, contested and wicked problems
A light plugin for investigating high-stakes, complex, contested, and wicked problems — the kind where the important things hide, the evidence is mixed or interested, and the deciding questions are value judgements. The engine surfaces the hidden contributors, conditions, and causal structure beneath a situation — sources of success and adaptation as well as failure — verifies every load-bearing claim against current sources, and routes what cannot be verified to your judgement. It offers understanding and a defensible next step, not control or prediction: wicked problems are navigated, not solved. The large language model provides the insight; the engine makes it verified, defensible, and audience-ready.
Built one controlled test at a time, keeping only what beats a strong plain pass. Every load-bearing claim is graded [V] (independently corroborated) or [N] (rests on an interested or self-reporting party), with strength tiers [V1]/[V2]/[V3]; once a grade is set it is never changed downstream — only re-expressed.
/analyse— the main workflow: a fail-safe proportionality triage → a fixed "provocation page" pass → source-tiered web verification with a disconfirmation search → a conditional systems-investigation pass for feedback-driven problems (a graded causal map of loops, tipping conditions, and leverage — investigate, don't predict) → deep-core routing → grade-lock + an assumption-resilience check → a forced judgement gate you clear → a five-part decision brief./verify— fact-check the load-bearing claims of an analysis against current sources; catch confidently-wrong facts; route what can't be verified online to the right place./render— re-voice a finished analysis for a specific reader (board, counsel, regulator, family, adversary, or any named standpoint) without changing a claim or a grade./track— keep an investigation live across sessions as new facts arrive.
It does not convene real stakeholders (simulated perspectives are labelled as such — participation is yours to arrange); it does not design interventions or trials (what-to-verify-first and flip conditions, yes; piloting and staging are routed to you); it does not simulate quantitatively (the systems pass grades structure and routes real modelling out); and it does not intervene in power — where evidence shows outcomes are power-determined, the brief says "power problem, not an analysis problem".
/plugin marketplace add ComplexAR/Insight-Engine
/plugin install insight-engine@insight-engine-marketplace
marketplace add takes owner/repo, so the command above assumes the repo is named Insight-Engine — substitute your repo name if you call it something else. The part after the @ is the marketplace's own name, set in .claude-plugin/marketplace.json. To pick up a new release later, run /plugin marketplace update.
The desktop app installs plugins from a file rather than from a GitHub marketplace:
- Download
dist/insight-engine-0.1.11.pluginfrom this repo (or from the latest Release). - In Cowork, open Customize → Skills → Upload from file and choose the
.plugin.
0.1.11. For how releases are cut, see PUBLISHING.md.
No license is set yet. Add a LICENSE file (for example MIT) before sharing widely, so others know the terms on which they can use it.